CBR for Tutoring and Help Systems

Abstract

Knowledge-based tutoring and help systems are designed to support learners and users individually. Such a capability stems from two so-called intelligent properties. Firstly, such a system is able to generate problem solutions automatically, based on its domain knowledge, without having to go back to inflexible, pre-compiled problem solutions (Self 1974). This allows the system to analyze complex and uncommon problem solutions, and to identify and explain errors. Secondly, representing the learners’ knowledge acquisition processes and their current knowledge in so-called learner models allows the system to adapt to the learners’ needs. Such learner models depend on representing the learners’ knowledge about the respective domain and on describing their ability to solve problems in that domain. In existing tutoring systems, these intelligent properties involve, to varying degrees, automatic, cognitive diagnoses of activities and problem solutions offered by the learner. Results from these diagnoses are used for supporting learners and for giving advice in a communication process (Wenger 1987).

Learner models based on overlays (Carr and Goldstein 1977) identify the learner's performance as a subset of an expert's capabilities. They are adapted in their view of explaining solutions or errors from a learner to the point of view of the expert who planned or programmed the system (Ohlsson 1986). However, even models using bug libraries (e.g., the Proust system, Johnson and Soloway 1987), or generative, runnable models (e.g, the program space approach, Ohlsson 1986), and the model-tracing method (Anderson et al. 1989) are limited in their ability to take into account the learners' intentions or their personal problem solving style.